The contribution of epigenetic alterations to natural variation for gene transcription levels remains unclear. In this study, we investigated the functional targets of the maize chromomethylase ZMET2 in multiple inbred lines to determine whether epigenetic changes conditioned by this chromomethylase are conserved or variable within the species. Gene expression microarrays were hybridized with RNA samples from the inbred lines B73 and Mo17 and from near-isogenic derivatives containing the loss-of-function allele zmet2-m1. A set of 126 genes that displayed statistically significant differential expression in zmet2 mutants relative to wild-type plants in at least one of the two genetic backgrounds was identified. Analysis of the transcript levels in both wild-type and mutant individuals revealed that only 10% of these genes were affected in zmet2 mutants in both B73 and Mo17 genetic backgrounds. Over 80% of the genes with expression patterns affected by zmet2 mutations display variation for gene expression between wild-type B73 and Mo17 plants. Further analysis was performed for 7 genes that were transcriptionally silent in wild-type B73, but expressed in B73 zmet2-m1, wild-type Mo17, and Mo17 zmet2-m1 lines. Mapping experiments confirmed that the expression differences in wild-type B73 relative to Mo17 inbreds for these genes were caused by cis-acting regulatory variation. Methylation-sensitive PCR and bisulfite sequencing demonstrated that for 5 of these genes the CpNpG methylation in the wild-type B73 genetic background was substantially decreased in the B73 zmet2-m1 mutant and in wild-type Mo17. A survey of eight maize inbreds reveals that each of these 5 genes exhibit transcriptionally silent and methylated states in some inbred lines and unmethylated, expressed states in other inbreds, providing evidence for natural variation in epigenetic states for some maize genes.
A variety of molecular mechanisms contribute to the phenotypic variation within a species. Allelic variation due to primary sequence variation in coding or regulatory regions has been widely documented (Buckler et al. 2006). Epigenetic variation, often associated with altered DNA methylation, histone modification, or chromatin structure, can also contribute to phenotypic variation. However, the relative importance of epigenetic variation in contributing to phenotypic variation is not well understood.
Epigenetic states can be quite stable and faithfully transmitted through mitosis and meiosis (Cubas et al. 1999; Chandler and Stam 2004). Epigenetic variation can result in formation of epigenetic alleles with altered gene expression patterns that can mimic traditional mutations. There are several examples of epigenetic variation that result in phenotypic variation (Bender and Fink 1995; Cubas et al. 1999; Morgan et al. 1999; Chandler et al. 2000; Stokes et al. 2002; Manning et al. 2006). Epigenetic variation has been classified as obligatory, facilitated, and pure according to the dependence on sequence changes (Richards 2006). Obligatory epigenetic variation requires a genetic change, such as a nearby transposon insertion, that conditions the allele to regulation by epigenetic mechanisms. Facilitated epigenetic variation requires a genetic change that conditions an allele to stochastically fall under epigenetic control, such as the intracisternal A-particle (IAP) insertion at some agouti alleles (Morgan et al. 1999). Pure epigenetic variation has no genetic changes associated with the epigenetic state. In many cases, the lack of detailed sequence comparisons makes it difficult to ascertain whether a particular instance is an example of an obligatory, facilitated, or pure epigenetic variation. Studies in Arabidopsis have provided evidence for natural variation for epigenetic states (Rangwala et al. 2006) and factors that effect epigenetic states (Riddle and Richards 2005), suggesting that epigenetic states may be a target for natural or artificial selection (Kalisz and Purugganan 2004).
DNA methylation is frequently associated with epigenetic alterations in higher plants. In plants, DNA methylation is most commonly found in CpG dinucleotide and CpNpG trinucleotide contexts, which are maintained by different enzymes (Chan et al. 2005). DNA methylation patterns are created and maintained by different families of DNA methyltransferase enzymes, including CpG maintenance methyltransferases (Finnegan and Dennis 1993; Kankel et al. 2003; Xiao et al. 2006), CpNpG maintenance/de novo methyltransferases (Henikoff and Comai 1998; Bartee et al. 2001), and de novo methyltransferases (Cao et al. 2000; Cao and Jacobsen 2002b). Additional factors, including a SWI/SNF2 factor (DDM1), a histone deacetylase (HDA6), a histone methyltransferase (KYP), and several components of the RNA interference silencing pathway, also play roles in maintaining proper DNA methylation levels (Vongs et al. 1993; Jeddeloh et al. 1999; Aufsatz et al. 2002; Jackson et al. 2002; Malagnac et al. 2002; Chan et al. 2006). Although unique targets of specific DNA methyltransferases have been identified, there is evidence for locus-specific, partial-to-complete redundancy for the function of DNA methyltransferases in the creation and maintenance of DNA methylation patterns (Cao and Jacobsen 2002a,b).
The plant-specific chromomethylase (CMT) enzymes contain a chromo-domain embedded within the methyltransferase motifs (Henikoff and Comai 1998; Genger et al. 1999; Bartee et al. 2001). The chromomethylases are primarily responsible for methylation found in CpNpG contexts, although they may also play a role in de novo methylation (Lindroth et al. 2001; Papa et al. 2001; Cao and Jacobsen 2002a,b; Cao et al. 2003). Although chromomethylases are important for silencing of endogenous sequences such as SUP and PAI loci (Bartee et al. 2001; Cao and Jacobsen 2002a), knockout mutations in CMT genes do not result in notable morphological phenotypes in Arabidopsis or maize (Bartee et al. 2001; Papa et al. 2001; Cao and Jacobsen 2002a). In addition to controlling the methylation state and expression level for SUP and PAI genes, CMT3 has been shown to methylate transposable elements (Lindroth et al. 2001). A series of genetic and biochemical studies have provided evidence for the targeting of CMT methylation through histone methylation (Lindroth et al. 2001; Jackson et al. 2002; Malagnac et al. 2002; Ebbs and Bender 2006) and short interfering RNA (siRNA) pathways (Zilberman et al. 2003; Chan et al. 2004, 2006).
In this study, we investigated the targets of the maize chromomethylase ZMET2 (Papa et al. 2001). Mutations in the maize Zmet2 chromomethylase gene result in a 10–15% reduction in genomewide methylation levels, mainly affecting CpNpG methylation (Papa et al. 2001). However, there is little evidence for any morphological phenotypes in homozygous mutant plants following multiple generations of introgression into several inbred lines (Papa et al. 2001). Surprisingly, in this study we found that ZMET2 was involved in the maintenance of epigenetic states that displayed natural variation in different maize inbreds. The expression and methylation status of epigenetically regulated alleles conditioned by CpNpG methylation was found to vary between different maize inbred lines, suggesting the potential for this epigenetic variation to contribute to transcriptional variation and phenotypic variation among inbred lines.
MATERIALS AND METHODS
Plant materials and tissue collection:
The zmet2-m1 allele, a loss-of-function allele resulting from an insertion of a Mu transposon into the 18th exon of the gene (Papa et al. 2001), was backcrossed into the Mo17, B73, and W22 r-g inbred lines for six generations and then self-pollinated for two generations to produce the homozygous near-isogenic lines of Mo17, B73, and W22 r-g that are homozygous for the zmet2-m1 allele. Biological replicates in the B73 background (six replicates) and Mo17 background (three replicates) were grown using standard greenhouse conditions; RNAs from these plants were used for microarray analyses. Each biological replicate was planted at a different time and represents the pooled aboveground tissue from eight 11-day-old seedlings (for full details on plant growth and tissue collection, see supporting online text at http://genetics.org/supplemental/). W22 r-g wild-type and zmet2-m1 plants, as well as plants of five other inbred lines (B14, B37, B84, Oh43, and Wf9) used for analysis of epigenetic variation in other inbreds, were grown under the same conditions. Plants segregating for the zmet2-m1 allele were genotyped with primers Zmet2F23, ZmetR23, and 9242 (see supplemental Table 1 at http://www.genetics.org/supplemental/ for all primer sequences) using the PCR conditions described below.
RNA isolation and microarray hybridization:
RNA isolation and Affymetrix (Santa Clara, CA) microarray hybridizations were performed as described (Stupar and Springer 2006) for six biological replicates of wild-type and mutant plants in the B73 genetic background and for three biological replicates of wild-type and mutant plants in the Mo17 genetic background. Spotted long oligonucleotide array hybridizations were performed for three biological replicates of the B73 and B73 zmet2-m1 genotypes using the same RNA stock used for Affymetrix hybridizations. For this experiment, dual channel (Cy3 and Cy5 per slide) studies were performed with complete dye swapping similar to previously described methods (Gardiner et al. 2005). Total RNA (1.5 μg) from each biological replicate was indirectly labeled using Ambion's (Austin, TX) Aminoallyl Message Amp II kit according to the manufacturer's recommendations with several modifications (see the supporting online text at http://www.genetics.org/supplemental/ for details of microarray hybridizations). The microarray data are available under GEO series accession no. GSE8188.
Microarray data analysis:
Affymetrix microarray data analysis was performed as described (Stupar and Springer 2006). Briefly, the GCOS software package v1.2 (Affymetrix) was used for signal acquisition and initial analysis. GeneSpring (Agilent Technologies, Palo Alto, CA) software was used for GC content robust multi-array analysis (GC-RMA) processing of the .cel files that involved normalization between the arrays and a subsequent per gene normalization of the resulting values (Stupar and Springer 2006). Genes differentially expressed in B73 zmet2-m1 relative to B73 or in Mo17 zmet2-m1 relative to Mo17 genotypes were identified by performing separate one-way ANOVA tests on the GC-RMA values using a parametric test with no assumption of equal variance for each genetic background. A Benjamin and Hochberg multiple testing correction was applied using a false-discovery-rate significance threshold of 0.25, such that 25% of the genes identified in a test are likely to be falsely identified. The genes that were identified in at least one of the comparisons were pooled together into a list of differentially expressed genes. The resulting list of the genes was further filtered on the basis of the criteria of expression level (i.e., the expression level must be >50 units for GC-RMA values in at least one of the genotypes) and significant t-test (P < 0.05 in B73 vs. B73 zmet2-m1, Mo17 vs. Mo17 zmet2-m1, or mutant vs. wild-type comparisons). The GeneSpring software was used to perform a hierarchical clustering analysis using a Pearson correlation method to create gene or condition trees based on specified gene lists, conditions, and genotypes.
For the long oligonucleotide spotted arrays, the data were extracted from each slide using GenePix Pro software (http://www.moleculardevices.com/pages/software/gn_genepix_pro.html). The analysis for differentially expressed genes was conducted using the limma package (Smyth 2004). The complete dye swapping was accounted for by determining the correlation values between the same samples coupled to both dyes. The data were normalized using loess and quantile methods to account for the variation within arrays and between arrays, respectively. Once the data had been background corrected and normalized and the dye bias had been accounted for, the data were fitted into a linear model and differentially expressed genes were estimated with a false discovery rate of 0.05 (Smyth 2004).
Primers were designed using Primer 3.0 software (Rozen and Skaletsky 2000) for a subset of the differentially expressed genes on the basis of corresponding tentative consensus contigs (http://www.tigr.org) and maize assembled genomic islands (Fu et al. 2005) and are listed in supplemental Table SOM1 at http://www.genetics.org/. RNA samples used for cDNA synthesis were DNAse treated (Promega, Madison, WI) and reverse transcribed using Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. PCR reactions were performed on genomic DNA or cDNA templates in a 15-μl total volume containing ∼50 ng of DNA, 2 pmol of each primer, 0.4 units of HotStar Taq polymerase (Eppendorf, Westbury, NY), 1.58 μl of 10× reaction buffer, and 0.1 μl of 50 mm dNTPs. Cycling conditions of the PCR reactions were as follows: 94° for 15 min, 35 cycles of 94° for 30 sec, 60° for 30 sec, 72° for 90 sec, followed by 72° for 5 min. Amplified products were separated in a 1% agarose TBE gel and visualized by ethidium bromide staining. The concentrations of cDNAs were calibrated and made approximately equal by amplifying a mez1 gene expressed at approximately the same level in all four genotypes analyzed.
Mapping of expression differences:
The genomic location responsible for expression differences for seven genes was mapped using a set of 51 recombinant inbred lines from the Syn4F7:9 IBM population (Lee et al. 2002). RNA isolation, cDNA synthesis, and RT–PCR reactions were performed as described above and each line was scored as “present” or “absent” for expression of the gene being amplified. These scores were mapped using “U Map It!” software (http://www.plantgenomics.iastate.edu/maize/).
Genomic DNA was extracted from 11-day-old seedlings for B73, B73 zmet2-m1, Mo17, and Mo17 zmet2-m1 genotypes using the CTAB method (Saghai-Maroof et al. 1984). Two plants were sampled for each of the four genotypes. For methylation-sensitive PCR assays, ∼500 ng of genomic DNA was digested with the methylation-sensitive enzymes PstI, DdeI, MspI, HpaII, or EcoRII according to the manufacturer's recommendations. The glycerol mock treatment was performed as a control for each of the DNA samples. The digested DNA was extracted by phenol–chloroform and precipitated with 0.1× vol sodium acetate (pH 5.5) and 2.5× vol ethanol. PCR reactions were performed on digested DNA templates as described above (see supplemental Table 1 at http://www.genetics.org/supplemental/ for all primer sequences). Bisulfite sequencing analysis was performed as described (Haun et al. 2007).
Annotations for differentially expressed genes were based on information available at the The Institute for Genomic Research Maize Gene Index (http://www.tigr.org/tigr-scripts/tgi/T_index.cgi?species=maize). The gene ontology (GO) annotations were obtained on the basis of assignment of the best Arabidopsis hits from the The Arabidopsis Information Resource website (http://www.arabidopsis.org/tools/bulk/go/index.jsp). The genetic map positions for Affymetrix array probe sets were predicted on the basis of identity with genetically mapped sequences or inferred on the basis of identity with BAC contig sequences that contained genetically mapped markers.
The expression of a limited number of genes is affected in zmet2 mutants:
To study the effects of disruption of CpNpG methylation in maize, we performed global profiling of gene expression levels in maize zmet2-m1 mutants using the Affymetrix maize GeneChip, which contains 17,622 probe sets corresponding to ∼13,500 genes. Although post-transcriptional regulation and protein modifications could affect the function of ZMET2 targets, in this study we investigated the gene transcript levels and used the word “expression” to refer to the abundance of RNA transcripts, not protein. Transcript levels of zmet2-m1 mutants and wild-type plants were compared in 11-day-old seedlings from two genetic backgrounds, B73 and Mo17, using separate statistical analyses of the GC-RMA processed signals from B73 and Mo17 genetic backgrounds. A series of statistical tests and expression criteria (see materials and methods) were employed to identify 126 differentially expressed genes (0.72% of all GeneChip probe sets) with 65 genes upregulated and 61 genes downregulated in zmet2-m1 mutants (Table 1; supplemental Table 2 at http://www.genetics.org/supplemental/). Many of the differentially expressed genes showed only a transcriptional response to the zmet2-m1 mutation in one of the two inbred backgrounds. Notably, one of the genes identified as downregulated in the zmet2-m1 mutant relative to wild type in both B73 and Mo17 was the zmet2 gene itself. The microarray results for 33 genes were validated by semiquantitative RT–PCR with 30 (91%) genes showing a strong correlation between microarray and RT–PCR results (Table 1; supplemental Table 2). The genetic map positions were predicted for ∼50% of the Affymetrix array probe sets based on identity with genetically mapped sequences or with BAC contig sequences containing genetically mapped markers. Analysis of the map positions for 52 of the 126 differentially expressed genes did not display any chromosomal bias (supplemental Table 2). Additionally, we did not detect evidence for allelic variants linked to the Zmet2 locus on chromosome 10 that may result in differential expression not dependent upon effects of the zmet2-m1 mutation.
Annotation of the differentially expressed genes:
The 126 differentially expressed genes were annotated with respect to putative function and inclusion in protein families. Of the 126 genes, 69 displayed significant similarity to rice or Arabidopsis genes with a putative function while 13 were similar to proteins with unknown function and 44 were not significantly related to any gene in rice or Arabidopsis (supplemental Table 2). To further analyze the functional roles of the ZMET2 targets, we obtained GO annotations for the genes affected in zmet2-m1 mutants on the basis of assignment of the best Arabidopsis hit. Of 126 genes affected by the zmet2 mutation, 66 displayed significant homology to an Arabidopsis gene and were used in GO analysis. We compared the relative representation of different GO categories for the differentially expressed genes vs. the total number of probe sets present on the Affymetrix arrays (supplemental Figure1 at http://www.genetics.org/supplemental/). We did not detect overrepresentation of any specific GO annotation in the genes affected in both B73 and Mo17 genetic backgrounds or in the genes affected in only one of the genetic backgrounds.
On the basis of studies from Arabidopsis that demonstrated a role for chromomethylases in maintaining DNA methylation of transposable elements, we might expect to see reactivation of silenced transposable elements in zmet2 mutant plants (Lindroth et al. 2001; Tompa et al. 2002; Kato et al. 2003). However, the Affymetrix maize microarray platform contains relatively few probe sets designed for repetitive features or transposable elements. Therefore, we hybridized the subset of RNA samples from B73 and B73 zmet2-m1 plants to a long oligonucleotide microarray platform with >50,000 features, including at least 355 probe sets derived from transposable elements (see supplemental online text, supplemental Table 3, and supplemental Figures 2 and 3 at http://www.genetics.org/supplemental/ for details on this experiment). Only 1 of the 179 differentially expressed genes identified using this platform was annotated as a transposable element (supplemental Table 3 at http://www.genetics.org/supplemental/), providing no evidence for extensive transcriptional activation of repetitive elements in the zmet2-m1 background. It is possible that the ZMET2 protein is involved in the methylation of repetitive elements in the maize genome (Papa et al. 2001) but that the transcriptional silencing of these elements is not solely dependent upon CpNpG methylation.
Complete loss of ZMET2 activity is required to cause changes in gene expression:
The altered DNA methylation phenotype of the zmet2-m1 mutant displays codominant inheritance on the basis of changes in CpNpG methylation at different genomic loci (Papa et al. 2001). To further validate the examples of differential expression and to determine if the altered expression patterns would exhibit codominant inheritance, we tested the expression of eight genes differentially expressed in B73 zmet2-m1 mutants in the sibling plants segregating for the zmet2-m1 allele (Figure 1). In all eight cases, regardless of whether the gene was up- or downregulated in the zmet2-m1 mutant, the heterozygous plants had the same expression levels as their wild-type siblings, suggesting that a complete loss of ZMET2 function is required for expression changes in these genes. The expression level of these eight genes was also tested in seedlings that were segregating for the zmet2-m2 allele, a second mutant allele of Zmet2 caused by a Mu insertion into exon 1 of the gene (data not shown). We found that the expression levels of these eight genes were similarly affected in plants homozygous for zmet2-m1 or zmet2-m2 alleles.
Comparison of expression profiles in different genetic backgrounds:
If zmet2-m1 regulates a set of conserved targets in maize, we would expect significant overlap of the differentially expressed genes identified in the B73 and Mo17 backgrounds. However, we found that the majority of differentially expressed genes (114/126) were differentially expressed in zmet2-m1 mutants in only one of the two genotypes (Table 1). The differentially expressed genes were classified into five groups on the basis of the expression levels in the four genotypes profiled (Table 2; Figure 2A). The group 1 genes were differentially expressed in zmet2-m1 mutants in both the B73 and the Mo17 genetic background. The group 2 genes showed differential expression in B73 but were expressed at equal levels in both genotypes in the Mo17 background while the group 3 genes were differentially expressed in B73 and were completely absent in the transcriptomes of both the mutant and the wild-type Mo17 genotypes. The last two groups of genes were examples of differential expression in Mo17 zmet2-m1 relative to Mo17 wild type while the transcripts of these genes in the B73 samples were either equally expressed in B73 (group 4) or completely absent in the B73 wild-type and mutant transcriptomes (group 5). Each of these five groups was subdivided according to whether the gene expression in the zmet2-m1 mutant plants relative to wild type was upregulated or downregulated. Representative examples of expression patterns observed for each of these gene classes and a clustering analysis of the affected genes based on microarray data are presented in supplemental Figure SOM4 at http://www.genetics.org/supplemental/.
One potential explanation for these differences between genetic backgrounds is that both genotypes display a similar trend in expression but that the expression changes were statistically significant in only one genetic background. However, analyses of the wild-type/mutant expression signal ratios (Figure 2B) suggested that the majority of these genes exhibit altered expression in zmet2-m1 mutants relative to wild type in one inbred but not in the other. We successfully validated the inbred-specific differential expression for 25 of the 29 genes tested by semiquantitative RT–PCR (Figure 3; Table 2).
It is possible that the differences in gene expression in the B73 and Mo17 are due to differences in gene content (Fu and Dooner 2002). We used PCR to test whether 29 of the genes that exhibited inbred-specific differential expression are present in genomic DNA of both inbreds (see supplemental Table 1 at http://www.genetics.org/supplemental/). We found that the majority of the genes that exhibited inbred-specific differential expression (28/29) are present in the genomic DNA of both B73 and Mo17. The one gene that was not detected in this assay (BQ703720 in Figure 3) is still clearly detected in the cDNA of Mo17 zmet2-m1 mutants, suggesting that the gene is present in both inbreds but that there is a sequence change in the Mo17 allele that interferes with amplification from the genomic DNA of Mo17. In addition, we performed BLAST searches against B73 genome survey sequences or BAC sequences to test for the presence of all of the 46 genes that exhibited differential expression specifically in the Mo17 genetic background. We found that each of the 46 genes that exhibited Mo17-specific transcriptional response is present in the B73 genome.
To further assess the prevalence of inbred-variable differential expression conditioned by ZMET2 among maize inbred lines, we compared the expression of 27 genes in wild-type and mutant plants in a third genetic background, W22 (Table 2; Figure 4). All four group 1 genes that were tested in W22 zmet2-m1 plants displayed similar expression change patterns in all three genetic backgrounds. Of the 23 genes from groups 2–5, only 4 showed evidence of differential expression in the zmet2-m1 mutant relative to wild-type plants in W22. This finding suggests that the majority of the genes from these groups (19/23) were affected in only one of the three genotypes tested. The widespread prevalence of differential effects in different genetic backgrounds suggested that there are few conserved targets of ZMET2 (likely genes from group 1). Instead, the genes regulated by ZMET2 may represent genes that independently fell under control of ZMET2 in different genetic backgrounds.
Analysis of inbred-specific Zmet2-conditioned epigenetic variation:
Our findings of gene expression patterns affected in zmet2-m1 mutants in some genetic backgrounds and not affected in others suggested the existence of natural variation for regulation of genes conditioned by ZMET2. Moreover, most of the ZMET2-affected genes showed large variations in expression level in B73 and Mo17 wild-type plants. Although most of the 126 genes that displayed altered expression in zmet2-m1 mutants are likely present in the genomes of both B73 and Mo17 inbreds, as evident from PCR analysis (data not shown), 93 (74%) exhibited differential expression between wild-type B73 and Mo17, with 43 (34%) showing no transcription in one of the two nonmutant genotypes (supplemental Figure 5 at http://www.genetics.org/supplemental/). To investigate the causes of natural variation for the epigenetic states conditioned by ZMET2, we selected 7 genes, herein referred to as chromomethylase-dependent expression (CDE) genes, for more detailed analysis. These 7 genes were completely silenced in wild-type B73 plants but were upregulated in B73 zmet2-m1 mutants. These 7 genes are expressed at comparable levels in both wild-type and zmet2-m1 mutant Mo17 plants (Table 3). We first assessed whether the expression variation in B73 relative to Mo17 was attributable to cis- or trans-acting regulatory factors through the use of recombinant inbred lines (see materials and methods). RT–PCR was used to assess whether or not expression was detected for these 7 CDE genes in 11-day-old seedling tissue harvested from 51 recombinant inbred lines from the intermated B73 × Mo17 population (Lee et al. 2002). For all 7 of the genes analyzed, the cause of the expression differences mapped to the same location as the gene itself, suggesting cis-regulatory variation (Table 3).
Since mutations in Zmet2 relieved the silencing of the B73 allele for these seven genes, it is likely that the regulatory variation of these genes is at least partially dependent upon epigenetic mechanisms. We were able to find partially assembled BAC sequences or genome shotgun sequences derived from B73 that included promoter regions for five of these seven genes (CDE1–5). To rule out the possibility that large-scale differences in the promoter regions of these genes between B73 and Mo17 inbreds were responsible for the expression differences, we amplified and sequenced portions of the promoter regions (1–4 kb) for these genes from Mo17 (Figure 5, A–E). For CDE1, CDE2, CDE3, and CDE5 we found SNPs but no evidence for large-scale rearrangements or insertions in the B73 allele relative to the Mo17 allele within the regions tested. An internal region of the CDE4 promoter was not amplified in Mo17 (Figure 5D). This could reflect structural differences in the promoter region or mismatches within the primer targets between Mo17 and B73 alleles.
The availability of genomic sequence for the promoter region of these five genes (CDE1–5) allowed us to test for methylation differences associated with altered expression levels in B73 and Mo17 inbred lines. Using methyl-sensitive PCR, we showed that, for all five of these genes, CpNpG sites were hypermethylated in wild-type B73 and hypomethylated in B73 zmet2-m1 and wild-type Mo17 (Figure 5, A–E), suggesting that the silent state in wild-type B73 was correlated with CpNpG methylation. We also screened three genes that displayed decreased expression in zmet2-m1 (likely secondary targets of ZMET2; supplemental Table SOM2 at http://www.genetics.org/supplemental/) and did not detect any methylation changes in any of the restriction sites tested. Bisulfite sequencing was used to confirm the CpNpG methylation differences within a 413-bp region of the CDE4 promoter (Figure 6). In the B73, B73 zmet2-m1, and Mo17 genotypes, there was a high level of CpG methylation. The B73 genotype displayed a relatively high level of CpNpG methylation that was significantly reduced in both B73 zmet2-m1 and Mo17 genotypes. While none of the cytosines studied displayed 100% methylation in B73 genomic DNA, there was consistent methylation at each of the CpNpG sites studied (Figure 6). Low levels of methylation were observed at cytosines in an asymmetric context in both B73 and Mo17 while the B73 zmet2-m1 plants did not exhibit cytosine methylation in an asymmetric context (Figure 6).
To determine whether the gene silencing conditioned by CpNpG methylation for ZMET2 targets is widely distributed among inbred lines or restricted to a single line, we tested the expression levels and promoter methylation states for CDE1–5 genes in seedling tissue from eight maize inbreds (Figure 5F; Table 3). For each of the genes we found that when the gene was expressed it was unmethylated and when the gene was not expressed there was CpNpG methylation at the restriction site(s) tested. Each of the five genes that we tested existed in a methylated and silenced state in at least two different inbred lines. One of the genes, CDE3, is methylated and silent in five of the eight inbreds that were surveyed. This finding, in combination with the ability to map the expression differences in recombinant inbred lines, suggests that the methylated and silenced allelic state of CDE genes is relatively stable.
DNA methylation patterns in plants are the result of a complex process. In addition to maintenance (MET) and domains rearranged methyltransferase (DRM) DNA methyltransferases, plants possess chromomethylases, responsible for maintenance of CpNpG methylation, which is specific to the plant kingdom. Chromomethylases are highly conserved across a variety of plant species and, therefore, should be functionally important. However, the exact role of CpNpG methylation has not been determined. Previous studies in maize (Papa et al. 2001) and Arabidopsis (Bartee et al. 2001; Lindroth et al. 2001; Cao and Jacobsen 2002a) have found little evidence for morphological phenotypes in CMT mutants. While mutations in either cmt3 or drm2 have very limited morphological phenotypes in Arabidopsis, the double mutant displays severe abnormalities, including developmental retardation and partial sterility, suggesting that, at least in Arabidopsis, chromomethylases may have partially redundant functions with the DRM methylases (Lindroth et al. 2001; Cao and Jacobsen 2002a,b). It is likely that partial redundancy also exists in maize. However, despite such functional redundancy, CMT3 has specific targets; for example, mutations in the cmt3 gene can have profound effects on the expression of the epigenetically controlled alleles of PAI or SUP (Bartee et al. 2001; Lindroth et al. 2001; Cao and Jacobsen 2002a).
Functional targets of the maize chromomethylase ZMET2:
To determine the functional targets of CpNpG methylation in maize, we used two different genomewide gene expression profiling platforms. We found that relatively few genes displayed altered expression. These genes are distributed throughout the genome and do not display marked trends in their putative function. Moreover, many of the genes affected by the zmet2 mutation display large differences in expression level in different maize inbred lines with some of the genes being completely silent in some inbreds, suggesting that the functional targets of ZMET2 are primarily nonessential genes that can exhibit altered expression levels without severe phenotypes. We found very little evidence for upregulation of transposable elements in zmet2 mutant plants. In Arabidopsis, the silencing of some transposable elements was shown to be a redundant function of CMT3 and MET1 (Kato et al. 2003) Therefore, it is possible that ZMET2 does play a role in the methylation of transposable elements but that the silencing of these elements is redundant with methylation catalyzed by other enzymes, preventing reactivation of mobile elements in maize zmet2 mutants.
One caveat in identifying the functional targets of ZMET2 is that a gene with altered expression may represent a primary target of ZMET2 (a gene that is directly methylated by ZMET2) or a secondary target (a gene that is regulated by a primary target). In the five genes that we tested, we found evidence for direct CpNpG methylation changes in promoter or intron sequences correlating to gene expression levels, but we expect that the set of 126 genes identified represents a combination of primary and secondary targets of ZMET2. The genes upregulated in zmet2 mutants when CpNpG methylation is released are more likely to represent primary targets of ZMET2. The finding that expression changes occur only in homozygous mutant individuals suggests that ZMET2 affects individual gene targets in a qualitative, rather than quantitative, manner.
Natural variation for the functional targets of ZMET2:
It is notable that the majority of functional targets of ZMET2 identified in our study represent genotype-specific targets. It is possible that there are additional conserved ZMET2 targets that were not identified in our study. Our inability to detect such conserved ZMET2 targets may be explained by the redundancy among CMT, MET, and DRM methyltransferases in regulation of the conserved functional targets of ZMET2. Therefore, the analysis of plants containing mutations in several methyltransferases of different types is required to identify such targets. Alternatively, our study of whole seedlings might have missed important cases of conserved regulation by ZMET2 that occur in specific tissues or at specific developmental points.
Initially, we were quite surprised by the intraspecific variation for the targets of ZMET2. However, after extensive validation of this variation and reconsideration of previous studies, this finding is not as surprising. The previously documented targets of CMT in Arabidopsis are PAI and SUP. Both of these cases represent examples of epigenetic states that are present in some ecotypes but absent in others. Similarly, many of the maize genes affected by ZMET2 identified in this study also displayed epigenetic states that appear to be variable among inbred lines. We found that multiple instances of differential expression in the zmet2-m1 plants in B73 or Mo17 genetic backgrounds represent genes that are present in the genome but not expressed at detectable levels in one of the inbreds despite the absence of major rearrangements within coding or regulatory regions.
The seven CDE genes represent examples of putative epigenetic variation due to the dependence of their expression state upon ZMET2, the cis-regulatory variation, and methylation changes associated with expression states. However, we cannot determine whether these represent potential obligatory, facilitated, or pure epigenetic variation. We do not have enough sequence from the B73 and Mo17 alleles to assess contributions of linked genetic changes. There is a high level of intraspecific variation for the content and identity of repetitive sequences and transposons surrounding maize genes (Fu and Dooner 2002; Brunner et al. 2005). We did not find evidence for variation in the content of transposable elements in the vicinity of the epigenetically controlled alleles identified in this study. It is likely that, at least for some of the alleles, alterations for the presence of specific repetitive sequences and/or transposable elements cause genes to become susceptible to regulation by ZMET2 in certain genetic backgrounds.
The ability to map the basis of the B73–Mo17 regulatory variation in recombinant inbred lines suggests that the expression differences exhibited by the B73 and Mo17 alleles are quite stable. In addition, it suggests that the expression differences cannot be communicated in a paramutation-like manner. If the silenced CDE allele were able to heritably alter the expressed CDE allele, we would not have been able to map the expression variation. Interestingly, if the allelic variation was caused by siRNAs, we might expect that the expressed CDE allele would be silenced when it is heterozygous with a silenced CDE allele. However, we did not find evidence supporting this mechanism. Moreover, we have evidence that one of the CDE genes exhibits mono-allelic expression in hybrids (R. M. Stupar and N. M. Springer, unpublished results). The presence of silenced alleles in multiple inbred lines suggests that such silent allelic states are stable and naturally exist within the population. The existence of naturally occurring epigenetically regulated alleles within maize suggests that these epigenetic changes may be selected upon and contribute to phenotypic variation. Naturally occurring epigenetic variation has been shown to produce different plant phenotypes and can have important implications for phenotypic variation and microevolution in natural plant populations (Kalisz and Purugganan 2004). Thus, ZMET2 participates in creating novel epigenetic variation through modulation of gene expression. Further experimentation is necessary to understand the mechanisms that produce epigenetically regulated alleles and their stability within maize lines.
The authors gratefully acknowledge the assistance of Peter Hermanson in assisting with plant growth, nucleic acid isolations, and validation PCRs. We are also grateful to several anonymous reviews for their helpful comments on this manuscript. This work was supported by grants from the National Science Foundation DBI-0227310 to N.M.S. and DBI-0321663 to S.M.K.
- Received February 27, 2007.
- Accepted July 18, 2007.
- Copyright © 2007 by the Genetics Society of America